The impact of climatic factors on headache patterns: a 14-year time series analysis of cluster and tension-type headaches in primary care

2.1 Data source and study population

A retrospective cohort study was conducted using data extracted from the electronic health records of patients from three primary care centers: “El Abajón” in Las Rozas, “Cerro del Aire” in Majadahonda, and “San Juan de la Cruz” in Pozuelo de Alarcón, Spain. The study period was from 01 January 2010 to 31 December 2023, and included all patients aged 18 years and older who consulted for headache-related conditions during the study period. Diagnoses were identified using the International Classification of Primary Care, Second Edition (ICPC-2), using the following codes: N89 (unspecified headache); N01 (tension-type headache); and N03 (cluster headache). No additional subcodes (N02/N04) existed in ICPC-2; therefore, only the confirmed codes recorded in the clinical system were used.

Sociodemographic variables, including age and sex, were retrieved from electronic medical records along with the consultation date. Clinical variables included headache subtypes.

Meteorological data were obtained from the Spanish Meteorological Agency (AEMET) weather station at Pozuelo de Alarcón. The AEMET station in Pozuelo de Alarcón is located approximately 15 km from Las Rozas de Madrid and 12 km from Majadahonda, Spain. The three health areas included in this study cover a total area of 140 km² (ID3194Y, latitude 40°26′54′′N, longitude 3°48′48′′W).

The study protocol was approved by the Research Ethics Committee of the Puerta de Hierro Majadahonda Hospital (PI 70/24, Act 06/2024). All procedures complied with the principles of the Declaration of Helsinki and ensured patient anonymity and confidentiality throughout the study. The study was conducted in accordance with the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) Statement and checklist [24].

All patients were included if they:

(1) were aged ≥18 years;

(2) had at least one primary care consultation registered between 01 January 2010 and 31 December 2023; and

(3) had a diagnosis coded with ICPC-2 N01 (tension-type headache), N03 (cluster headache), or N89 (unspecified headache).

Patients were excluded if:

(1) their medical record lacked age or sex information;

(2) the consultation corresponded to duplicate entries; or

(3) the diagnostic code did not belong to the headache-related ICPC-2 categories defined for this study.

2.2 Climatic variables

Meteorological data corresponding to the same geographical region and study period were obtained from the Spanish State Meteorological Agency (AEMET). The variables collected included mean temperature (°C), diurnal temperature range (°C), day-to-day temperature change (°C), wind direction (°), mean wind speed and wind gust speed (m/s), daily barometric pressure range and day-to-day barometric pressure change (hPa), and sunshine hours (h/day). Because wind direction is a circular variable (0° and 360° representing the same direction), it cannot be directly modeled as a linear predictor. Therefore, the wind direction was transformed into sine and cosine components, which allowed circular data to be appropriately represented in the regression models.

Daily meteorological data were aggregated into weekly values to match the frequency of medical consultations. These meteorological indicators were chosen based on prior studies suggesting that atmospheric changes, particularly temperature, barometric pressure, and humidity, may influence neurological symptoms and pain modulation [13, 19, 20, 25].

2.3 Follow-up process for consultations

To capture longitudinal patterns in headache-related healthcare utilization, the number of primary care consultations associated with headache diagnoses was aggregated on a weekly basis; however, individual patient trajectories were not analyzed. This longitudinal follow-up enabled the identification of repeated medical visits for each headache subtype (CH, TTH, and unspecified headache), thereby allowing the analysis of temporal fluctuations and recurrence patterns in primary care.

2.4 Sample size

According to Burmeister et al. [26], at least 240 observations are required to ensure sufficient power for a multiple regression model with 11 continuous predictors and one dichotomous predictor. The present study substantially exceeded this requirement, including 5127 consultations recorded between 2010 and 2023. Weekly data aggregation produced a time series of approximately 728 observations, allowing robust estimation of both the ETSX and ARIMAX models with external regressors. This calculation is provided for orientation only, as traditional power calculations do not directly apply to time-series models such as ETSX or ARIMAX.

2.5 Study procedures

Patient data, including age, sex, and ICPC-2 diagnostic codes, were anonymized and extracted from electronic medical records. These clinical variables were systematically linked to the corresponding meteorological parameters obtained from the Spanish State Meteorological Agency (AEMET) for the same observation period. The climatic dataset included mean temperature (°C), diurnal temperature range (°C), day-to-day temperature change (°C), wind direction (°) transformed into cosine and sine wind directions, mean and gust speed (m/s), daily barometric pressure range and its day-to-day change (hPa), and sunshine hours (h/day) [27].

Each variable was aggregated weekly to align with the frequency of medical consultations and ensure consistency between environmental exposure and health care activities.

2.6 Outcome measures

The primary outcome was the weekly number of primary care consultations for headaches, subdivided into tension-type, cluster, and unspecified headaches, as recorded using ICPC-2 diagnostic codes.

The secondary outcomes included the evaluation of temporal trends, sex- and age-related differences, and the influence of climatic variables on consultation frequency. Diagnostic criteria were consistently applied across 14-year period, and meteorological variables were defined according to the AEMET operational standards to ensure homogeneity and reproducibility across all data sources.

2.7 Statistical analysis

Statistical analysis was performed using R version 4.1.3 (R Foundation for Statistical Computing, Institute for Statistics and Mathematics, Welthandelsplatz 1, 1020 Vienna, Austria).

The significance level was set at p < 0.05.

Quantitative variables are described as mean ± standard deviation, and qualitative variables are presented as absolute and relative frequencies (%).

The number of cases was analyzed from 01 January 2010 to 31 December 2023 using a time-series analysis, with weekly counts of CH, unspecified headache, and TTH. External regressors included age, sex, mean temperature, diurnal temperature range, day-to-day temperature change, wind direction, mean wind speed, and gust speed, as well as daily barometric pressure range and its day-to-day change, after excluding variables with a variance inflation factor (VIF) greater than 10. In the case of wind direction, because it is a circular variable where North is both at 0° and 360°, it was segmented into the cosine and sine of the wind direction and represented using sine and cosine components to appropriately model circular data [28]. The model selection, Exponential Smoothing State Space Model with external regressors (ETSX), or AutoRegressive Integrated Moving Average with external regressors (ARIMAX), was carried out by analyzing the accuracy of the predictions of a training set (75% of the sample) on a test set (25% of the sample) and comparing the root mean squared error (RMSE), the Symmetric Mean absolute percentage error (SMAPE), and the mean absolute scaled error (MASE), in which the lower the value, the better, between both [29, 30] selecting the model with the best overall value across the three statistics. The stationarity of the series was tested using the Augmented Dickey-Fuller test, as well as compliance with the assumptions using the Ljung-Box test (autocorrelation of the residuals) and the Kolmogorov-Smirnov test with Lilliefors correction (normality of the residuals).

Due to the presence of 14 weeks in which there were missing data in the variables age, sex (male and female), CH, Headache, and TTH, which represents 1.92% of the series, missing data were imputed using the Predictive Mean Matching (PMM) method implemented in the mice package in R [31] to ensure analytical consistency. All statistical procedures were double-checked for reproducibility and internal consistency.

4.1 Principal findings

This 14-year time-series analysis, conducted to examine whether climatic variables are associated with weekly primary care consultations for cluster headache, tension-type headaches, and unspecified headaches, revealed that only a small subset of climatic variables showed independent associations with headache consultations. Moreover, the nature of these associations differed across the headache subtypes. The ARIMAX model best captured the dynamics of tension-type headache (TTH), whereas the ETSX model provided superior predictive accuracy for overall headache, unspecified headache, and cluster headache (CH), in agreement with the model comparison metrics from the supplementary analyses.

Among all headache types, sex was the most consistent predictor. In the TTH series, female sex was the only demographic factor significantly associated with an increased number of consultations, whereas male sex was not significant. Average temperature showed a modest but significant protective effect, reducing the weekly number of TTH consultations.

For CH, the male sex was significantly associated with reduced consultation frequency, whereas no climatic variable reached significance in the final ETSX model. In the overall and unspecified headache models, sex remained a significant regressor; however, the negative coefficients observed for both the female and male sex should be interpreted in relation to the model intercept and do not indicate a higher consultation burden in men. These findings are consistent with the descriptive data showing a predominance of female consultations. However, no meteorological variable showed an independent association. Overall, these results indicate that the influence of weather on headache consultations is weaker and more subtype-specific than traditionally assumed and is largely overshadowed by demographic factors, particularly sex.

4.2 Comparison with previous studies

The present findings refine and partially challenge previous evidence suggesting the strong influence of meteorological variables on headache occurrence [10, 11, 12, 13, 14, 15, 17]. Earlier studies have associated temperature variability, humidity, and barometric pressure changes with increased risk of migraine, TTH, or CH attacks [10, 11, 12, 13, 14, 15, 17]. However, many of these investigations relied on patient-reported triggers, smartphone-based symptom diaries, or short observational windows [19, 20], which may amplify the perceived relevance of weather-related factors.

Our results diverge from population studies indicating a strong climatic modulation of CH periods, particularly through temperature variation [13, 17], because temperature, wind, rainfall, pressure, and sunshine hours were not significant predictors of CH in our long-term primary care dataset. Similarly, previous work linking barometric pressure changes to TTH exacerbations [11, 32] was not corroborated by our models, in which pressure-related variables were consistently nonsignificant.

Sex-related findings in our models echo long-established epidemiological patterns: female sex was associated with higher TTH consultation rates, whereas male sex was not independently associated, in line with global burden estimates [8]. Conversely, the protective effect of male sex on CH aligns with contemporary CH epidemiology [25].

The disparity observed in this study between the number of men and women seeking consultations for headaches is likely partly related to sex differences in healthcare-seeking behaviors. Women have consistently been shown to engage more frequently in healthcare services when experiencing pain, particularly in primary care settings. Psychosocial factors, including heightened concern about symptoms, gender-related expectations, and emotional distress, may further increase the likelihood of women seeking medical attention for headache-related complaints. Recent population-based evidence indicates that women are more likely than men to seek healthcare across a wide range of symptoms and conditions, even after accounting for sociodemographic factors and health status. This supports the interpretation that consultation-based data may reflect behavioral patterns in healthcare utilization rather than true differences in disease occurrence alone [33].

Overall, the present study suggests that climatic factors exert a weaker influence on healthcare utilization for headaches than previously hypothesized, particularly when analyzed over long temporal frames using objective registry data [18, 19, 20, 22].

4.3 Mechanistic considerations

Although most climatic variables did not independently predict headache consultations, the few significant associations—temperature and wind direction for TTH—may reflect underlying neurophysiological pathways described in biometeorological research. Variations in ambient temperature can influence autonomic balance, vascular tone, and the thermoregulatory mechanisms involved in trigeminovascular activation [7, 10, 16]. Directional changes in wind flow are often accompanied by transitions in barometric pressure, humidity, and ionization patterns, which may modulate pericranial muscle tension or trigeminal afferent excitability [18, 22, 34].

The lack of climatic associations in CH contrasts with hypotheses involving hypothalamic sensitivity to environmental cues and circannual rhythmicity [21]. It is possible that individual-level triggers, such as sleep irregularities, stress, or behavioral changes, exert a much stronger influence on CH attacks than population-level meteorological patterns.

The post-2020 increase in overall headache consultations observed in the present study likely reflects pandemic-related psychosocial factors rather than meteorological effects, including increased stress, altered sleep patterns, and emotional distress, all of which are known contributors to headache frequency [35, 36]. In addition, increased healthcare utilization following the COVID-19 pandemic has been reported not only for headaches but also for other orofacial pain conditions [37, 38].

4.4 Clinical and public health implications

From a clinical standpoint, these findings provide a nuanced perspective on the role of weather in headache management. Although biometeorological influences exist, their magnitude appears modest and may not justify strict behavioral restrictions for most patients. Furthermore, from a clinical perspective, it is important to recognize that patients often attribute their headaches to weather-related factors; however, as shown in this study, climatic variables appear to play a more limited role than previously assumed, and patient education should therefore aim to help deconstruct such beliefs and expectations [23, 32]. Nevertheless, clinicians should incorporate climatic factors as secondary modulators, particularly in individuals who report personal meteorosensitivity, as suggested in the literature [23, 32].

The validated ETSX and ARIMAX models offer a foundation for developing forecasting tools capable of anticipating fluctuations in consultation demand, particularly given the strong demographic patterns observed. Such predictive models may help optimize primary care workload distribution, support patient education, and guide individualized self-management strategies during periods of abrupt weather changes.

However, the present results caution against overstating the epidemiological impact of weather factors on headache. Personalized profiling, as advocated in recent biometeorological frameworks [22], may better capture the clinically relevant subsets of patients sensitive to climatic variations.

In addition, adverse weather conditions may influence healthcare-seeking behavior, potentially delaying or discouraging primary care visits, independent of headache severity. As individual-level data on access patterns or delayed consultations were not available, this potential behavioral confounding factor could not be directly assessed and should be considered when interpreting consultation-based outcomes.

4.5 Limitations and future directions

Several limitations of this study must be acknowledged. First, reliance on primary care diagnostic codes may underrepresent true headache prevalence and may not fully capture clinical severity or attack characteristics. Second, meteorological data were obtained from a single AEMET station, limiting the assessment of microclimatic variability within the catchment area. Third, the study did not include potentially interacting environmental exposures, such as air pollution, allergens, or noise, some of which have been linked to increases in headache-related emergency visits [39, 40]. Fourth, the absence of patient-reported outcomes prevented us from examining subjective meteorosensitivity, attack timing, and individual trigger perceptions. Another limitation regarding climatic variables was the absence of humidity data, despite evidence in the literature suggesting that humidity may be associated with headache occurrence [3, 6, 10, 11, 12, 13, 14, 17, 18, 19, 20, 25]. An important methodological limitation of this study is the reliance on primary care consultation data as a proxy for headache occurrence. Not all headache episodes lead to a medical visit, particularly among individuals who self-manage their symptoms, experience recurrent or habitual headaches, or perceive limited benefit from repeated consultations. As a result, consultation counts may not fully reflect the true temporal distribution or frequency of headache episodes in the underlying population. This limitation is particularly relevant when interpreting associations with climatic variables, as the observed time-series patterns primarily represent healthcare-seeking behavior rather than the actual incidence of headache attacks.

Future studies should integrate multi-station meteorological data, incorporate pollutant measures, and combine registry-based models with smartphone-based symptom tracking. Machine learning approaches and individualized trigger profiling may substantially improve the predictive accuracy and clarify the mechanisms linking climate and headache exacerbations. In clinical practice, it is common for patients to report worsening of symptoms associated with cold or wind flow. It would be interesting to correlate these data with more specific diagnoses, such as cervicogenic headaches or headaches secondary to temporomandibular disorders.